compiling error
Posted: Fri Aug 31, 2007 5:32 pm
I found a problem in my VASP parallel calculations. I calculated a Al (110) slab by using 10 processors for k-points of 3X3X1. The calculation
went well. But when I tried to use 12 processors for the same structure,
I failed. I also tried to use 10 processors for 4X4X1 k-points, I failed
either.
I checked the OUTCAR file, the tail line is as follows:
> ..........................................................................
> .
> .
> .
>
> maximum number of plane-waves: 10660
> maximal index in each direction:
> IXMAX= 10 IYMAX= 15 IZMAX= 15
> IXMIN=-10 IYMIN=-15 IZMIN=-15
> NGX is ok and might be reduce to 42
> NGY is ok and might be reduce to 62
> NGZ is ok and might be reduce to 62
> real space projection operators:
> total allocation : 35860.19 KBytes
> max/ min on nodes : 35860.19 35860.19
>
> parallel 3dFFT wavefunction:
> minimum data exchange during FFTs selected (reduces bandwidth)
> parallel 3dFFT charge:
>
> /STOP HERE/ and /nothing follows !
> /...............................................................................................
>
> I recomplied the VASP,
> I noticed something as follows:
> ...............................................................................
> fft3dlib.f90(4763) : (col. 12) remark: loop was not vectorized:
> existence of vector dependence.
> fft3dlib.f90(4784) : (col. 12) remark: loop was not vectorized:
> vectorization possible but seems inefficient.
> fft3dlib.f90(4816) : (col. 9) remark: loop was not vectorized: not inner
> loop.
> fft3dlib.f90(4818) : (col. 12) remark: loop was not vectorized:
> vectorization possible but seems inefficient.
> fft3dlib.f90(4840) : (col. 9) remark: loop was not vectorized: not inner
> loop.
> fft3dlib.f90(4843) : (col. 25) remark: loop was not vectorized:
> unsupported loop structure.
> ............................................................................................................
>
> I need your help for this problem. thanks.
>
> Yushan
went well. But when I tried to use 12 processors for the same structure,
I failed. I also tried to use 10 processors for 4X4X1 k-points, I failed
either.
I checked the OUTCAR file, the tail line is as follows:
> ..........................................................................
> .
> .
> .
>
> maximum number of plane-waves: 10660
> maximal index in each direction:
> IXMAX= 10 IYMAX= 15 IZMAX= 15
> IXMIN=-10 IYMIN=-15 IZMIN=-15
> NGX is ok and might be reduce to 42
> NGY is ok and might be reduce to 62
> NGZ is ok and might be reduce to 62
> real space projection operators:
> total allocation : 35860.19 KBytes
> max/ min on nodes : 35860.19 35860.19
>
> parallel 3dFFT wavefunction:
> minimum data exchange during FFTs selected (reduces bandwidth)
> parallel 3dFFT charge:
>
> /STOP HERE/ and /nothing follows !
> /...............................................................................................
>
> I recomplied the VASP,
> I noticed something as follows:
> ...............................................................................
> fft3dlib.f90(4763) : (col. 12) remark: loop was not vectorized:
> existence of vector dependence.
> fft3dlib.f90(4784) : (col. 12) remark: loop was not vectorized:
> vectorization possible but seems inefficient.
> fft3dlib.f90(4816) : (col. 9) remark: loop was not vectorized: not inner
> loop.
> fft3dlib.f90(4818) : (col. 12) remark: loop was not vectorized:
> vectorization possible but seems inefficient.
> fft3dlib.f90(4840) : (col. 9) remark: loop was not vectorized: not inner
> loop.
> fft3dlib.f90(4843) : (col. 25) remark: loop was not vectorized:
> unsupported loop structure.
> ............................................................................................................
>
> I need your help for this problem. thanks.
>
> Yushan